2:00 PM - 2:15 PM
[SCG45-26] Variations of earthquake spectral content and their implications for earthquake nucleation
Keywords:Stress drop, Fluid injection, Spectral analysis, Machine learning approach
Earthquake source parameters provide fundamental knowledge about rupture mechanics. Their spatial and temporal variations reflect the stress and strength evolution within the fault zone, and are influenced by both intrinsic strength distribution and external triggering processes. However, the diversity in observed trend leads to challenges in interpreting those variations. In this study, we focus on spectral variations of fluid-triggered earthquakes in Oklahoma, where earthquakes of distinct frequency contents are identified using regional stations and dense nodal arrays.
In Cushing, Oklahoma, where a M5 earthquake occurred in 2016. We deployed a nodal array with 130 three-component seismic stations in 2019. During the 30-day period, there were 3 earthquakes recorded by the regional seismic network, 12 earthquakes from a Machine-learning (ML) catalog using regional stations. We detected and located over 100 microearthquakes by combining ML algorithm and beam-forming method. Applying unsupervised ML algorithm in spectral clustering, we found that the events occurred on the main fault (hosted the M5 earthquake) have substantially different spectral content than the other newly detected faults. Events on the new faults are enriched in low-frequency content compared to on-main-fault earthquakes. To better understand whether such low-frequency content is due to source or path effect, we apply spectral ratio analysis using the empirical green’s function (EGF) method, and found lower stress drop for these off-fault earthquakes than events on the main fault.
From 2019-2022, an isolated sequence occurred in southeast Oklahoma near Quinton related to hydraulic fracturing and wastewater injection. Applying the unsupervised ML algorithm a regional seismic station that recorded the whole sequence, we identify four distinct clusters with different frequency contents. Similarly, we apply spectral ratio analysis to separate source and path effects. We find that low-frequency cluster have very low stress drop (~1 kPa), comparable to slow slip events, while the M3 earthquakes exhibiting normal stress drop values (~3 MPa). Those low-stress drop earthquakes occurred prior to the M3 earthquakes and disappeared after the M3 events. We will validate the stress drop results using a dense nodal array deployed in 2021 with a few earthquakes occurred during the one-month period.
Combining the results from the two sequences, we hypothesize that there exists strong evidence of slow-slipping faults in the fluid-injection seismic zone, and there also exist temporal interactions between slow-slip and regular seismic events.
In Cushing, Oklahoma, where a M5 earthquake occurred in 2016. We deployed a nodal array with 130 three-component seismic stations in 2019. During the 30-day period, there were 3 earthquakes recorded by the regional seismic network, 12 earthquakes from a Machine-learning (ML) catalog using regional stations. We detected and located over 100 microearthquakes by combining ML algorithm and beam-forming method. Applying unsupervised ML algorithm in spectral clustering, we found that the events occurred on the main fault (hosted the M5 earthquake) have substantially different spectral content than the other newly detected faults. Events on the new faults are enriched in low-frequency content compared to on-main-fault earthquakes. To better understand whether such low-frequency content is due to source or path effect, we apply spectral ratio analysis using the empirical green’s function (EGF) method, and found lower stress drop for these off-fault earthquakes than events on the main fault.
From 2019-2022, an isolated sequence occurred in southeast Oklahoma near Quinton related to hydraulic fracturing and wastewater injection. Applying the unsupervised ML algorithm a regional seismic station that recorded the whole sequence, we identify four distinct clusters with different frequency contents. Similarly, we apply spectral ratio analysis to separate source and path effects. We find that low-frequency cluster have very low stress drop (~1 kPa), comparable to slow slip events, while the M3 earthquakes exhibiting normal stress drop values (~3 MPa). Those low-stress drop earthquakes occurred prior to the M3 earthquakes and disappeared after the M3 events. We will validate the stress drop results using a dense nodal array deployed in 2021 with a few earthquakes occurred during the one-month period.
Combining the results from the two sequences, we hypothesize that there exists strong evidence of slow-slipping faults in the fluid-injection seismic zone, and there also exist temporal interactions between slow-slip and regular seismic events.